<aside> 📢
Paid partnership with Fish Audio.
</aside>
<aside> 🎬
This is how I made the voices in the AI drama from the reel, in order: setup → building a character voice → directing Drama 3 → the real job (fixing one line in an AI video).
The directions below are the exact ones I used in my drama "REUNION". The lines are Korean because the drama is Korean; directions work in other languages too.
</aside>
Open the link below and hit Sign up at the top right.
👉 https://fish.audio/?fpr=hyo-eun60
| What you want | Where |
|---|---|
| Turn lines into speech | Text to Speech (fish.audio/tts) |
| Pick an existing voice | Voice Library (2,000,000+ voices) |
| Make a character voice from sample audio | Voice Cloning (fish.audio/app/voice-cloning) |
| Make a new voice from a written description | Voice Design (fish.audio/app/create-voice) |
| API key (only if you run it from code) | fish.audio/app/api-keys |
In Text to Speech, pick Drama 3 as the model version. The Try Drama 3 button on the Fish Audio home page opens it with Drama 3 already selected. It's still in Preview, and Fish's own page says results vary between takes, so generate a few and pick the best.
Once you have 10+ lines and want to compare several tones, clicking one at a time in the browser gets slow. From that point I hooked the API up to Claude Code.
fish.audio/app/api-keys.envFISH_API_KEY=your_key
Use the Fish Audio API to generate this line with the drama-3-preview model.
Put the [acting direction] in front of the line, save 6 separate takes, and let me listen to them side by side.
<aside> 💡
The model value I used in the API was drama-3-preview. Fish's page also shows drama-3 in one place, so check the API docs for the current value before you run it. If the model value is wrong, it can quietly fall back to another model without an error.
</aside>